Data set IIa

Group Information:
 Department of Biomedical Engineering, 
 Tsinghua University, 
 Beijing, 100084, P. R. China
Supervisor:  GAO Xiaorong, Ph.D.  ( gxr-dea@tsinghua.edu.cn )
Team Member: CHENG Ming,  JIA Wenyan
Counselor:   GAO Shangkai, YANG Fusheng


Algorithm description

The algorithm includes three steps: spatial filtering, feature
extraction[GS1], and classification. A proper selection of spatial
filtering method could increase the signal-to-noise ratio and improve
the classification accuracy. Two features are extracted respectively
in power spectrum and time domain after preprocessing, and a
2-dimentional linear classifier is adopted for final classification.

Spatial filtering
Spatial filtering includes common average reference (CAR) and common
spatial subspace decomposition (CSSD). The CAR method can be
implemented easily and we reserve the method in our preprocessing. The
CSSD method is designed to compare the activation of brain response
components that are differently activated in two cognitive task
conditions, and extract signal components which are specific to one
condition compared to another condition.

Power feature extraction
Denote as the power spectral density of channel .  is the set of the
selected channels and is the set containing all the frequency bins
within the frequency band of mu rhythm, the power feature is defined
as


Time feature extraction
Denote as the time course of mu rhythm of channel , obtained by narrow
band filter. The energy accumulation function of channel is defined as

Denote  as the set of the selected channels. The time feature is defined as


Linear classifier
One point is drawn in the plane for each trial whose coordinates are
the two features. Three lines divide the plane into four regions, and
each region represents one target.  [GS1]